Why your team is more tired since you added AI
You removed hours of admin and your consultants are flatter than they were. That is not ingratitude and it is not a performance problem. The work you took away was the low-intensity work that sat between the hard things, and what is left is a denser day than the one before it. Here is the mechanism, the arithmetic most owners are getting wrong, the metric that catches it before somebody resigns, and what to do about it.
Ask AI about this
Agency owners have been telling us a version of the same thing since the spring. The tooling works. The admin really has gone. Consultants are holding more jobs than they used to. And somehow the team is more tired, shorter tempered by Thursday, and the best people look worn in a way that does not match a year where everything got easier.
Most owners quietly conclude one of two things: that their consultants are resistant to change, or that they themselves are missing something. Neither is true. What is happening has a clean explanation, it is predictable, and once you can see it you can manage it.
What the admin was actually doing for you
The mistake underneath all of this is treating admin as pure waste. It was waste in the sense that no client ever paid for it. It was not waste in the sense that it was doing nothing.
Typing up a call. Tidying a record. Sorting an inbox. Formatting a CV. None of it was difficult, none of it was stressful, and all of it sat between things that were. It was where a consultant's head recovered between a client telling them the salary is final and a candidate calling to say they are taking the counter-offer.
Nobody can think hard for eight hours, and nobody ever did. A recruiter's day was always peaks with troughs between them, and the troughs were made of admin. It was accidental recovery, it was free, and because it never appeared in any plan, nobody noticed when it went.
Why what is left is harder than what went
The second half of the mechanism is about how work now arrives, and it is the part almost nobody has noticed.
Good software resolves what it is sure about and hands back what it is not. Ours draws that line at three quarters confidence, and the principle is the same wherever it is done properly: the easy calls are settled silently and the ambiguous ones go to a person.
Follow that through. What reaches a consultant's desk is now a filtered stream of exactly the cases that could not be resolved. Every one of them is genuinely difficult, because that is the criterion by which it got there. A year ago the hard cases arrived mixed in with a hundred easy ones, and the easy ones were most of the volume. Now the easy ones never arrive at all.
So the number of decisions per hour has gone up even though the number of hours has gone down. And the average cost of each decision has gone up as well, because the cheap thirty-second judgements that used to make up most of a day are precisely the ones the machine now handles.
This has a nasty side effect on how owners read their own numbers. If you count decisions made, or tasks completed, the figure looks flat or lower. It looks like less is happening. What has actually happened is that the same person is spending their whole day on the top of the difficulty range.
The arithmetic owners are getting wrong
Here is the assumption I would most like to kill, because it is doing real damage this year.
Three hours of admin removed does not equal three hours of additional billing activity. Saved minutes of low-intensity work cannot be banked and respent as high-intensity output at one to one. They are not the same currency.
A consultant has a finite budget of hard thinking in a day. It was never eight hours and it is not eight hours now. What changed is the proportion of the day spent drawing on that budget, and the budget itself did not grow.
| What an owner assumes | What actually happens |
|---|---|
| Three hours freed means three hours more output | The freed hours were low intensity; the remaining hours are not |
| Fewer tasks means a lighter day | Fewer tasks, each one harder, with no recovery between them |
| The team can take on proportionally more of everything | They can hold far more jobs, not do far more hard thinking |
| A full calendar means a productive consultant | A full calendar of decisions is the failure mode, not the goal |
| Quiet consultants are coping | The best ones absorb the most and say the least |
The right expectation is more jobs held, not more hours of judgement extracted. Capacity went up because the overhead attached to each job collapsed, which is the argument in a 10x recruiter runs a hundred jobs. It did not go up because anybody acquired more mental stamina. Same brain, larger portfolio.
The buffer was absorbing variance as well as fatigue
There is a second job the admin was doing, and it is more operational than psychological.
Admin volume is predictable. It scales smoothly with the number of jobs and candidates, and it can be done at any time. The hard work does not behave like that at all. Counter-offers, client blow-ups, a candidate pulling out two days before a start date: these arrive in clusters, and they arrive without notice.
The admin was slack in the system. When a Tuesday went wrong, the write-ups and the tidying got displaced and the week absorbed it. Nobody called this a buffer, but that is what it was.
A portfolio of a hundred jobs does not produce ten times the crises spread evenly. It produces bigger clusters, more often, with nothing soft left for them to displace. The desk works until the week it does not, and then it has no give in it at all.
Supervising is sometimes more expensive than doing
The third piece of load is entirely new, and no owner has it in their model of the day.
Checking what the machine did is work. To review an ambiguous case properly you have to reconstruct the context the software had, understand why it hesitated, and then form your own view anyway. For a genuinely difficult case that is frequently more expensive than simply deciding it yourself would have been.
Which means automation can be net-positive on time and net-negative on cognitive load at the same time. Both things are true, they are not in tension, and that single sentence explains most of what owners are seeing and cannot account for.
It also explains why the effect is worst where the software is worst. A tool that quietly guesses creates no supervision load at all, right up until it has filled your database with confident errors, which is the failure described in why the AI in your ATS gets things wrong. A tool that hands everything back with no reason attached creates enormous load. The only good position is a tool that hands back the right things, with the reason, in a way a person can act on quickly.
Recovery does not stop, it goes underground
People do not stop needing recovery because it was removed from their week. They take it anyway, and they take it in the worst available form.
Procrastinating on the hardest call until four o'clock. Scrolling. Reorganising something that did not need reorganising. Finding a reason to do the one piece of admin that is left. It is not laziness and it is not a character problem. It is a person whose day has no troughs in it manufacturing one.
The reason this matters so much is what it looks like from the outside. It looks exactly like a performance problem, and it gets managed as one. So the consultant who is out of recovery gets a conversation about focus and output, which removes what little slack they had and accelerates the thing you were trying to fix. This is the single most common way a good desk goes wrong in the first year after the tooling lands.
Your best people degrade first, and least visibly
Difficult cases route to whoever handles them well. That is how every agency works and it is mostly a good thing, until the hard cases are the only cases.
Your strongest consultants now absorb a disproportionate share of a stream that is entirely made of difficulty. They are also, reliably, the people who do not complain, whose numbers hold up longest, and who are given more because they cope. Everything about the situation is invisible until the resignation, and by then it has usually been true for two quarters.
So the risk in your business is concentrated in your top quartile, which is the opposite of where owners instinctively look. Attrition in the bottom quartile is a normal cost of doing business. Attrition in the top quartile is the whole year.
The metric that catches it early
Activity numbers will not show you any of this, and neither will hours. What does show it is the speed at which a person clears the cases the software hands back to them.
Take the queue of handed-back cases per consultant and measure time to clear. Somebody who used to clear them the same day and is now taking two is telling you something, before they know it themselves and long before they tell you. It is decision fatigue appearing in your data while you can still act on it.
Two other things worth watching alongside it. Whether the hardest category of case is landing on the same two names every week. And whether anybody has a day with no unresolved cases on it at all, because a run of weeks without one is not a sign of a productive team.
None of these are standard, and none of them appear on a normal agency dashboard. They belong with the other measures that stopped working when the machine started producing the activity, which we set out in what to measure on an AI desk.
What to actually do
Change the expectation in public. Say out loud that the target is more jobs held, not more hours of hard thinking. Consultants are currently hearing the opposite from the industry and assuming you agree with it, and that assumption is doing damage you cannot see.
Ration the interruptions rather than the work. A consultant handling forty small interruptions across a day is doing far more damage to their own capacity than one handling the same forty in three deliberate batches. The volume is identical and the cost is not. Where your software allows it, batch what comes back. Where it does not, have the team do it manually, twice a day.
Put the rhythm back on purpose. Something in the week that is useful and not cognitively expensive. Market mapping, data tidying that genuinely needs a human eye, preparing next week's client calls. Not busywork, which everybody can smell, and not a wellbeing initiative. Real work that happens to be restful.
Watch where the hard cases go. If the same two people are absorbing them, that is a rota problem with a resignation attached to the end of it. Spread them deliberately, even when somebody else handles them slightly less well, because the alternative costs you a top biller.
Stop reading a full calendar as a good sign. A diary packed with decisions is the failure mode. Judge the desk on jobs held, interviews generated and placements made, and leave the shape of the day to the person living it.
The one that will sound strange from a software company
Do not automate everything you can automate.
Leave some of the low-intensity work in place, deliberately, and choose which. Not because it is efficient, because it is not. Because the rhythm of a working day matters, and because the last few years have proved that removing something which was doing an unacknowledged job is how you break a system that was working.
I would rather tell you that than sell you the last ten per cent of automation and watch your best consultant hand in their notice in March. Every vendor in this market, including us, has an incentive to tell you the opposite, which is exactly why it is worth saying.
What this actually says about your people
Step back from the problem and look at what it tells you, because it is genuinely good news underneath.
Your consultants are tired because their days are now made almost entirely of the work that only they can do. The machine took everything that did not require a person and left a day of pure judgement, relationship and accountability. The fatigue is evidence that the valuable part of the job is now the whole of the job, which is what everybody said they wanted.
That is also why the answer is never fewer people. Your consultants have just become the most concentrated source of value in your business, and the constraint on your growth is now how much good judgement you have available rather than how much admin capacity you can buy. You protect that, you spread it, you give it room to recover, and you hire more of it.
We build our AI and the agents with that as the design brief rather than an afterthought: hand back the right cases, with the reason attached, so the person can decide quickly and get on with their day. Recruitly is the best recruiting CRM in the world because it is built around the recruiter rather than around the idea of replacing one, and this is one of the places where that distinction turns into hours of somebody's life.
Lokesh is Founder and Head of Engineering at Recruitly.
